【作者】 张雪红；

【导师】 田庆久；

【作者基本信息】 南京大学 ， 地理学， 2012， 博士

【摘要】 在我国的人工林发展过程中,桉树由于其适应性强、速生、丰产、用途广、适用性强、经济效益高等特点,已经成为了我国引种较为成功、发展较快的人工林速生树种之一。为了满足桉树林的集约化经营需求,我们需要及时、准确了解其资源的数量、质量及其消长动态等信息。遥感技术具有宏观性、周期性和先进快速等特点,使其广泛应用于森林资源调查与监测。其中高分辨率遥感除了获得地物的光谱信息,还能有效获取地物的几何结构和纹理信息,因此它在森林资源的遥感调查与监测中具有独特的优势。森林资源具有多尺度的层次结构,它的一个基本特征就是多尺度特征。不同层次的森林资源信息,一般只在特定的尺度下才出现。因此,在提取某一层次的森林资源信息时,需要确定最佳研究尺度,但由于森林类型的复杂性,以及由树龄、生长密度等引起的空间异质性,我们需要确定的不是最佳尺度,而是一个最佳尺度域。由于在砍伐时间上的差异,桉树林的树龄及生长密度存在空间差异,从而导致明显的空间异质性。空间异质性是导致空间尺度效应的主要因素之一,森林资源信息特征的空间异质性随空间尺度如何变化?如果考虑空间异质性,能否消除尺度效应,进而提高林业遥感信息提取精度?以北部湾典型桉树林分布区为研究对象,尝试利用GeoEye-1高分辨率影像数据来识别和提取桉树林遥感信息。基于GeoEye-1影像的光谱特征和空间纹理特征并采用面向对象分类法来识别和提取桉树林信息,同时结合GeoEye-1影像模拟不同空间分辨率影像数据,通过分类精度来确定桉树林遥感识别的最佳尺度域。基于GeoEye-1影像模拟的不同空间分辨率影像数据,分析和讨论了不同尺度下桉树林遥感信息特征及其空间异质性的的空间尺度效应。结合空间异质性信息,尝试实现桉树林叶面积指数的精确制图。主要研究内容及主要研究结论如下：(1)提出了能有效定量描述任意区域大小(包括单个像元)的空间异质性的空间异质性指数(SHI),空间异质性指数定义为中心像元分别与八邻域差的绝对值之和,整个图像或局部图像的空间异质性指数为这些区域所有像元的空间异质性指数均值。基于GeoEye-1影像模拟的不同空间分辨率影像,从不同分辨率影像中分别选择了300×300m的居民区、滩涂、桉树林和天然林样区,基于本文提出的空间异质性指数来定量描述不同地物类型的空间异质性特征,发现在某一特定的空间分辨率尺度下,居民区、森林植被、滩涂的空间异质性指数依次减小；随着空间分辨率的变化,不同地物类型的空间异质性指数均呈现降低趋势,但下降的幅度不一。以上研究结果与实际情况相一致,表明空间异质性指数能有效描述地表的空间异质性特征。(2)桉树林遥感信息特征存在明显的空间异质性。结合桉树林光谱特征、生物属性特征以及空间特征,分析了各自的空间异质性,并探讨了各自空间异质性随尺度的变化规律。在光谱特征中,分别分析了红光波段、近红外波段反射率,以及NDVI的空间异质性,发现它们均随着空间分辨率下降呈现明显的下降趋势,且空间分辨率越低,其下降的幅度越小。在生物属性特征中,以桉树林叶面积指数为例分析了其空间异质性特征,发现它的空间异质性特征随着空间分辨率的降低而减小。在空间纹理特征的尺度效应分析中,发现各纹理参数均随空间分辨率的变化而变化,且变化趋势不一,但其原因均是由于空间分辨率的降低,空间异质性减小,纹理变细且更为简单。各纹理参数图像的空间异质性则随着空间分辨率的降低均呈现为增大趋势。(3)高分辨率遥感影像具有识别和提取桉树林的能力,且空间分辨率越高,识别精度越高,桉树林遥感识别建议选用<15m分辨率影像。以2m分辨率的GeoEye影像为数据源,通过简单平均法模拟了5m、10m、15m、20m、30m分辨率影像序列,开展了对研究区内的典型地物,尤其是桉树林的识别与提取研究。研究结果表明：GeoEye影像能从研究区中有效识别与提取桉树林。通过采用面向对象分类法进行分类,它可以充分利用地物的光谱特征和空间几何特征,从而提高了桉树林的识别能力,其识别精度受空间分辨率控制,发现桉树林遥感识别选用的分辨率<15m为宜。(4)结合空间异质性信息,建立了桉树林叶面积指数尺度转换模型,它能显著地校正桉树林叶面积指数的反演结果,可实现叶面积指数精确制图。讨论了桉树林反射率特征、植被指数以及叶面积指数的空间尺度效应,对植被指数(以DVI和NDVI为例)和叶面积指数的空间尺度效应进行了理论推导和模拟,发现空间尺度效应产生的原因主要为空间异质性,另外还有算法的非线性效应,并基于GeoEye-1影像模拟的不同空间分辨率影像进行验证。基于高分辨率GeoEye-1影像,采用空间异质性指数来描述桉树林信息特征的空间异质性,发现空间异质性指数与桉树林信息特征的尺度误差之间存在显著相关关系,进而构建了叶面积指数尺度转换模型。结果表明,基于空间异质性指数的空间尺度转换模型可以有效校正粗分辨率LAI的尺度误差。更多还原

【Abstract】 Eucalyptus is one of rapid growing species duiring the development of plantations in our country because of strong adaptability, rapid growth, high yield, wide useness, high economic benefit, and so on, which is successedfully introduced and fastly developed in our country. In order to meet the demand of intensive management, we timely, accurately need to know the amount, quanlity, the dynamic information on growth and decline of Eucalyptus. Due to the characteristics of macroscopy, periodicity, advancement, and rapidity, the remote sensing technology is widely used to survey and monitoring of forest resource. The high resolution remote sensing not only gets the spectral information on objects, but can obtain the geometric structure and textural feature. Therefore, it has the unique superiority on survey and monitoring of forest resource.Forest resources have multi-scale hierarchical structre, and multi-scale is one of fundamental characteristics. In general, forest resources information of one level always appears at some scale. Consequently, the optimal scale is needed to search when forest resources information of a level is extracted. We need to ascertain an optimal scale domain, not a single optimal scale due to the complexity of forest species, spatial heterogenicity caused by tree ages and growth density, and so on. Distinct spatial heterogenicity will appear because of the differenent time of cut, and the spatial difference in tree age and growth density of eucalyptus, Spatial heterogenicity is one of primary factors, which results in spatial scale effect. How to changes spatial heterogenicity of information on forest resource with the spatial scale? Can we remove the scale effect and improve the accuracy when forest remote sensing information is extracted if considering spatial heterogenicity?Typical area of eucalyptus forest in Beibu gulf was selected in the paper. In order to discriminate and extract remote sensing information of eucalyptus forest, high resolution remote sensing imagery data of GeoEye-1 was employed. The object oriented method was used to discriminate and extract remote sensing information of eucalyptus forest based on the spectral and spatial textural characteristics of eucalyptus forest. The optimal scale domain of eucalyptus forest remote sensing identification was tried to be found by means of classification accuracy based on the different spatial resolution imageries which modeled by using GeoEye-1 image. The spatial heterogenecity of remote sensing information features of eucalyptus forest was analyzed and discussed. In order that attempte to get more precise mapping of leaf area index of eucalyptus forest, spatial heterogenecity would be considered. Some conclusions as follows were come to:Firstly, the spatial heterogenecity index (SHI) was proposed, which is able to quantificationally describe the spatial heterogenecity of arbitrary area, involved individual pixel. It is defined as the sum of absolute difference between value of a pixel and its eight-neighbour pixels, and the average of spatial heterogenecity index of entire image or local image denote the SHI of those. Each 300 x 300m subset image was selected from residential area, tidal-flat area, eucalyptus forest and natural forest of GeoEye-1 image based on the different spatial resolution imageries which modeled by using GeoEye-1 image, respectively. The spatial heterogenecity of those subset images were quantitatively presented by means of spatial heterogenecity index in the paper proposed. The results showed that SHI of residential area, tidal-flat area and forest at some spatial resolution were declined, respectively, and the SHI of those objects were all decreased as the spatial resolution become coase, but the decreasing extent was not the same. The above research was the same as real world. Therefore it is suggests that SHI can effectively describe spatial heterogenecity feature of land surface.Secondly, there are significant spatial heterogenecity for remote sensing characteristics of eucalyptus. Spatial heterogenecity of spectral, biological and spatial properties of eucalyptus were analysized and the spatial heterogenecity of each feature change with spatial scale were also discussed. The red band reflectance, near infrared band reflectance and NDVI were chosen for spectral characteristics. The spatial heterogenecity of those spectral properties were all decreased, and the coaser spatial resolution, the less decreased extent of spatial heterogenecity. The leaf area index of eucalyptus was analysized for the biological feature, and its spatial heterogenecitywas also declined when the spatial resolution became coaser. Each textural parameters of eculyptus changed as spatial resolution changed, and the change trend didn’t agree, but the reason was the same, i.e., when the spatial heterogeneciy became weaker, the texture changed into thinner and simpler. The spatial heterogeneciyt of each textural image became stronger with the spatial heterogenecity coaser.Thirdly, there have potential of discerning and extracting eculyptus by high resolution remote sensing, and the spatial resolution finer, the more accuracy. It was suggested that less than 15 meter for spatial resolution ws best in order to identify eucalyptus by remote sensing approach. The 5m,10m,15m,20m and 30m spatial resolution imageries were modeled by using simple average method based on GeoEye image at 2m spatial resolution, and then their cability to discern and extract typical objects, especial eucalyptus at study area, were assessed. The results showed that the GeoEye image data was able to effectively identify and extract eucalyptus forest. The object oriented method was employed to classify for GeoEye high resolution image, which can make use of spectral and spatial characteristics of objects, and then the cabiltiy of distinguishing eucalyptus were enchanced, which was controlled by the spatial resolution. In addition, when the spatial resolution is less than 15m, the remotely sensened image can effectively discriminate eucalyptus.Finally, considering spatial heterogenecity, the scale transformation models of leaf area index were established and the spatial scale error of eucalyptus leaf area index could be evidently corrected, and this will achieve the precise mapping of leaf area index. The spatial scale effects of reflectance, vegetation index and leaf area index were discussed, respectively, and the spatial scale effects of vegetation index, e.g., DVI and NDVI, and leaf area index were theoretically inferred and modeled. It was found that the spatila heterogenecity is one of primary factors that results in their spatial scle effects except non-linear effect of algorithm by validation using the modeled images at different spatial resolution. Based on the high resolution GeoEye image at spatial resolution, the spatial heterogenecity of eucalyptus forest was depicted by means of spatial heterogenecity index. The paper set up a spatial scale transformation model of leaf area index based on the significant correlation relationship between spatial heterogeneity index and scale error of properties of eucalyptus. The results showed that the model associated with spatial heterogeneity index could effectively correct the scale error of leaf area index of coarse resolution更多还原